Checkout: suggestions for large repository

[softprops]

I'm experimenting with Github Actions on a large code repository at $WORK. We use a mix of Concourse ci and Jenkins and are looking towards Github Actions as a potential CI/CD tool with less hosting maintenance.

In some simple experiments we found the actions/checkout command to be slower that expected. A checkout step takes ~6m 53s using with.fetch-depth: 1, and that's before my job can do anything useful.

In our Jenkins set up we have a persistent clone that we do local clones from. In concourse we use the git resource. In both cases the process of fetching a given version of the code feels faster than actions/checkout and was wondering if there as any actions tuning parameters would could apply to speed up the process

I can see a glimpse into what the action is doing in the output.

git remote add origin https://github.com/{org}/{repo}
 git config gc.auto 0
git config --get-all http.https://github.com/{org}/{repo}.extraheader
git config --get-all http.proxy
git -c http.extraheader="AUTHORIZATION: basic ***" fetch --tags --prune --progress --no-recurse-submodules --depth=1 origin +refs/heads/*:refs/remotes/origin/*

Is there a way we can opt out of features like fetching all of the tags (we have a lot of tags) or submodules (we don't use submodules)?

I started looking at want concourse git resource does to draw some comparisons. Looking here I can see it does something like a clone + checkout operation

git clone --single-branch $depthflag $uri $branchflag $destination $tagflag

cd $destination

git fetch origin refs/notes/*:refs/notes/* $tagflag

if [ "$depth" -gt 0 ]; then
  "$bin_dir"/deepen_shallow_clone_until_ref_is_found_then_check_out "$depth" "$ref" "$tagflag"
else
  git checkout -q "$ref"
fi

I'm not sure if that's any better or worse our experience is that it "feels" more performance that what we see with github actions/checkout.

[zoispag]

I have similar problem.

git fetch + git checkout takes ~4 minutes (TeamCity needs a few seconds, because it only applies the changes from previous run)

I solved it with:

steps:
    - name: Clone working branch
      run: git clone --single-branch --branch ${{ github.head_ref }} --depth 1 https://${{ secrets.CLONE_TOKEN }}:[email protected]/${{ github.repository }}.git .

Shallow clone takes ~ 30 seconds.

Maybe we need an official shallow-clone action?

[softprops]

@zoispag just curious what was the scope applied to secrets. CLONE_TOKEN vs the default GITHUB_TOKEN that GitHub provides to each action?

[zoispag]

@softprops I gave repo access.

Because I needed to clone via https from a private repo, I used the OAuth key of an actual account/github user

[softprops]

Thanks. It would be great if GITHUB_TOKEN had access to do that. I feel hesitant about these personal access tokens with repo status because they give access to all of the repos your github user has access to. I believe GITHUB_TOKEN provided by GitHub is scoped to only permissions for the specific repo.

In any case I tried your solution and it worked! down from 7 minutes I'm at about 1m 22s. That's a huge savings. Thank for for posting a reply and sorry I was late to notice it!

[zoispag]

No worries for late replying!
I feel the same about the CLONE_TOKEN, but GITHUB_TOKEN did not allow to perform this task. 😞Though I am using a "dedicated" GitHub account, which generally acts as a bot, so by definition has access only to this repo.

7m to 1m22s is a huge improvement, so happy to help! 🎉 😁

[softprops]

Codified what worked for us here
https://github.com/meetup/express-checkout

[zoispag]

@softprops I wanted to do the same, but didn't have the time yet.
It would be great to include me as a contributor in your README though 😉

[softprops]

@zoispag done! https://github.com/meetup/express-checkout/pull/2

[XhmikosR]

This is definitely something that needs to be improved. Fetching all tags and branches can be slow.

[ericsciple]

fixing this in v2, will hopefully merge early next week

[ericsciple]

@softprops @XhmikosR checkout v2-beta (now in master). waiting for feedback/stabilization, then will push v2 tag. v2 fetches only a single commit by default

[stayallive]

@ericsciple I've enabled it in our repo and we went from 30-40 seconds checkouts to 5-7 seconds ❤️ (even in our teeny tiny 12k commits repo) this is gut! I can only imagine what actual bug repo's will see as improvements!

Thanks for all the work from you and everyone involved!

[softprops]

I haven't had a change to take a look yet but would it be possible to post a comment to this issue when there's an official v2 release published?

[softprops]

I just ran a benchmark on our largest repo. This now comparable in performance to our work around. As soon as the beta version changes to v2 we're just going to switch to actions/checkout proper

[ericsciple]

published v2 tag

[softprops]

Works great! Thanks folks

[gregwym]

with checkout@v2 it indeed saves a lot of bandwidth but seems much slower on the download speed. Please see the comparison bellow. I suspect it is because GitHub has to do the compression on the fly when fetching with --depth=1?

V1

V2